COMPASS and ACANet accepted by Nature Medicine and Nature Communications, respectively.
FPPool, a knowledge-guided pooling framework connecting molecular fingerprints and graph neural networks, is on ChemRxiv.
Medea, an omics AI agent for automated therapeutic discovery, is posted on bioRxiv.
Dr. Shen joins Zhejiang University as Assistant Professor under the Hundred Talents Program.
AI Agents for Drug Discovery: Closing the Loop from Hypothesis to Molecule — Drug discovery is slow, expensive, and iterative. AI agents — systems that reason, plan, and use tools autonomously — are poised to close the design–synthesize–test loop at unprecedented speed. We discuss architecture, challenges, and our work on Medea and Leadmaster.
FPPool: Bridging Molecular Fingerprints and Graph Neural Networks — We introduce FPPool, a knowledge-guided graph pooling framework that uses chemical fingerprints to decompose molecular graphs into biochemically meaningful substructures. State-of-the-art on 40 benchmarks with built-in interpretability.
MolMap: Transforming Molecular Fingerprints into 2D Feature Maps — MolMap converts molecular descriptors into structured 2D feature maps, enabling CNNs to learn spatial patterns in molecular property prediction with state-of-the-art accuracy. Published in Nature Machine Intelligence.